Optimization of Classifiers Performance for Node Embedding on Graph Based Data

Abstract

The Publications regarding the research of embedding the large-scale information that helps in getting networks utilizing neighborhood-aware node representations and low-dimensional communities cover a wide area of research. In graph mining applications, these classification models, and embedding performed better than the conventional approaches. When using different conventional machine learning and data analysis approaches, the display of graphs and their relationship is highly useful in describing features present. Many different embedding approaches are used in machine learning, and a literature review was conducted to determine the best techniques for comparison. This study examines the accuracy scores of different classifiers using the approach on a single dataset. The dataset which is used in this study is CORA, and it is used to import it. After the network has been formed using the dataset, the nodes are embedded since the result of this node embedding will be used as a training set. The machine learns through training of model, for which the Node2vex method is applied in this work. The classifiers are used to train the model. Gradient Boosting, Logistic Regression, Random Forest, K-Neighbors, Decision Tree, Gaussian, and SVC are the classifiers utilized to solve this model's classification problem. To assess performance, the model makes use of two classifiers: Gradient Boosting, Logistic Regression, Random Forest, K-Neighbors, Decision Tree, Gaussian, and SVC. Through experimentation, the accuracy score is used to compare the classifier’s levels of efficiency. From the study, it was clearly observed that for the dataset, it was only the Support Vector Classifier that performed best in the testing and training of dataset for getting desired result. This was achieved by achieving an accuracy of 0.7706 and an MCC score of 0.7200. The optimum classifier for model training tasks and node classification can be chosen with the aid of this paper.